net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50 #include <linux/bvec.h>
  51 #include <linux/highmem.h>
  52 #include <linux/uio.h>
  53 #include <linux/sched/mm.h>
  54
  55 #include <trace/events/sunrpc.h>
  56
  57 #include "sunrpc.h"
  58
  59 static void xs_close(struct rpc_xprt *xprt);
  60 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  61                 struct socket *sock);
  62
  63 /*
  64  * xprtsock tunables
  65  */
  66 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  67 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  68 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  69
  70 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  71 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  72
  73 #define XS_TCP_LINGER_TO        (15U * HZ)
  74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  75
  76 /*
  77  * We can register our own files under /proc/sys/sunrpc by
  78  * calling register_sysctl_table() again.  The files in that
  79  * directory become the union of all files registered there.
  80  *
  81  * We simply need to make sure that we don't collide with
  82  * someone else's file names!
  83  */
  84
  85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  90
  91 static struct ctl_table_header *sunrpc_table_header;
  92
  93 /*
  94  * FIXME: changing the UDP slot table size should also resize the UDP
  95  *        socket buffers for existing UDP transports
  96  */
  97 static struct ctl_table xs_tunables_table[] = {
  98         {
  99                 .procname       = "udp_slot_table_entries",
 100                 .data           = &xprt_udp_slot_table_entries,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = proc_dointvec_minmax,
 104                 .extra1         = &min_slot_table_size,
 105                 .extra2         = &max_slot_table_size
 106         },
 107         {
 108                 .procname       = "tcp_slot_table_entries",
 109                 .data           = &xprt_tcp_slot_table_entries,
 110                 .maxlen         = sizeof(unsigned int),
 111                 .mode           = 0644,
 112                 .proc_handler   = proc_dointvec_minmax,
 113                 .extra1         = &min_slot_table_size,
 114                 .extra2         = &max_slot_table_size
 115         },
 116         {
 117                 .procname       = "tcp_max_slot_table_entries",
 118                 .data           = &xprt_max_tcp_slot_table_entries,
 119                 .maxlen         = sizeof(unsigned int),
 120                 .mode           = 0644,
 121                 .proc_handler   = proc_dointvec_minmax,
 122                 .extra1         = &min_slot_table_size,
 123                 .extra2         = &max_tcp_slot_table_limit
 124         },
 125         {
 126                 .procname       = "min_resvport",
 127                 .data           = &xprt_min_resvport,
 128                 .maxlen         = sizeof(unsigned int),
 129                 .mode           = 0644,
 130                 .proc_handler   = proc_dointvec_minmax,
 131                 .extra1         = &xprt_min_resvport_limit,
 132                 .extra2         = &xprt_max_resvport_limit
 133         },
 134         {
 135                 .procname       = "max_resvport",
 136                 .data           = &xprt_max_resvport,
 137                 .maxlen         = sizeof(unsigned int),
 138                 .mode           = 0644,
 139                 .proc_handler   = proc_dointvec_minmax,
 140                 .extra1         = &xprt_min_resvport_limit,
 141                 .extra2         = &xprt_max_resvport_limit
 142         },
 143         {
 144                 .procname       = "tcp_fin_timeout",
 145                 .data           = &xs_tcp_fin_timeout,
 146                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 147                 .mode           = 0644,
 148                 .proc_handler   = proc_dointvec_jiffies,
 149         },
 150         { },
 151 };
 152
 153 static struct ctl_table sunrpc_table[] = {
 154         {
 155                 .procname       = "sunrpc",
 156                 .mode           = 0555,
 157                 .child          = xs_tunables_table
 158         },
 159         { },
 160 };
 161
 162 /*
 163  * Wait duration for a reply from the RPC portmapper.
 164  */
 165 #define XS_BIND_TO              (60U * HZ)
 166
 167 /*
 168  * Delay if a UDP socket connect error occurs.  This is most likely some
 169  * kind of resource problem on the local host.
 170  */
 171 #define XS_UDP_REEST_TO         (2U * HZ)
 172
 173 /*
 174  * The reestablish timeout allows clients to delay for a bit before attempting
 175  * to reconnect to a server that just dropped our connection.
 176  *
 177  * We implement an exponential backoff when trying to reestablish a TCP
 178  * transport connection with the server.  Some servers like to drop a TCP
 179  * connection when they are overworked, so we start with a short timeout and
 180  * increase over time if the server is down or not responding.
 181  */
 182 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 183
 184 /*
 185  * TCP idle timeout; client drops the transport socket if it is idle
 186  * for this long.  Note that we also timeout UDP sockets to prevent
 187  * holding port numbers when there is no RPC traffic.
 188  */
 189 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 190
 191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 192 # undef  RPC_DEBUG_DATA
 193 # define RPCDBG_FACILITY        RPCDBG_TRANS
 194 #endif
 195
 196 #ifdef RPC_DEBUG_DATA
 197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 198 {
 199         u8 *buf = (u8 *) packet;
 200         int j;
 201
 202         dprintk("RPC:       %s\n", msg);
 203         for (j = 0; j < count && j < 128; j += 4) {
 204                 if (!(j & 31)) {
 205                         if (j)
 206                                 dprintk("\n");
 207                         dprintk("0x%04x ", j);
 208                 }
 209                 dprintk("%02x%02x%02x%02x ",
 210                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 211         }
 212         dprintk("\n");
 213 }
 214 #else
 215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 216 {
 217         /* NOP */
 218 }
 219 #endif
 220
 221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 222 {
 223         return (struct rpc_xprt *) sk->sk_user_data;
 224 }
 225
 226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 227 {
 228         return (struct sockaddr *) &xprt->addr;
 229 }
 230
 231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 232 {
 233         return (struct sockaddr_un *) &xprt->addr;
 234 }
 235
 236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 237 {
 238         return (struct sockaddr_in *) &xprt->addr;
 239 }
 240
 241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 242 {
 243         return (struct sockaddr_in6 *) &xprt->addr;
 244 }
 245
 246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 247 {
 248         struct sockaddr *sap = xs_addr(xprt);
 249         struct sockaddr_in6 *sin6;
 250         struct sockaddr_in *sin;
 251         struct sockaddr_un *sun;
 252         char buf[128];
 253
 254         switch (sap->sa_family) {
 255         case AF_LOCAL:
 256                 sun = xs_addr_un(xprt);
 257                 strlcpy(buf, sun->sun_path, sizeof(buf));
 258                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 259                                                 kstrdup(buf, GFP_KERNEL);
 260                 break;
 261         case AF_INET:
 262                 (void)rpc_ntop(sap, buf, sizeof(buf));
 263                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 264                                                 kstrdup(buf, GFP_KERNEL);
 265                 sin = xs_addr_in(xprt);
 266                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 267                 break;
 268         case AF_INET6:
 269                 (void)rpc_ntop(sap, buf, sizeof(buf));
 270                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 271                                                 kstrdup(buf, GFP_KERNEL);
 272                 sin6 = xs_addr_in6(xprt);
 273                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 274                 break;
 275         default:
 276                 BUG();
 277         }
 278
 279         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 280 }
 281
 282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 283 {
 284         struct sockaddr *sap = xs_addr(xprt);
 285         char buf[128];
 286
 287         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 288         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 289
 290         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 291         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 292 }
 293
 294 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 295                                      const char *protocol,
 296                                      const char *netid)
 297 {
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 300         xs_format_common_peer_addresses(xprt);
 301         xs_format_common_peer_ports(xprt);
 302 }
 303
 304 static void xs_update_peer_port(struct rpc_xprt *xprt)
 305 {
 306         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 307         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 308
 309         xs_format_common_peer_ports(xprt);
 310 }
 311
 312 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 313 {
 314         unsigned int i;
 315
 316         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 317                 switch (i) {
 318                 case RPC_DISPLAY_PROTO:
 319                 case RPC_DISPLAY_NETID:
 320                         continue;
 321                 default:
 322                         kfree(xprt->address_strings[i]);
 323                 }
 324 }
 325
 326 static size_t
 327 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 328 {
 329         size_t i,n;
 330
 331         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 332                 return want;
 333         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 334         for (i = 0; i < n; i++) {
 335                 if (buf->pages[i])
 336                         continue;
 337                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 338                 if (!buf->pages[i]) {
 339                         i *= PAGE_SIZE;
 340                         return i > buf->page_base ? i - buf->page_base : 0;
 341                 }
 342         }
 343         return want;
 344 }
 345
 346 static ssize_t
 347 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 348 {
 349         ssize_t ret;
 350         if (seek != 0)
 351                 iov_iter_advance(&msg->msg_iter, seek);
 352         ret = sock_recvmsg(sock, msg, flags);
 353         return ret > 0 ? ret + seek : ret;
 354 }
 355
 356 static ssize_t
 357 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 358                 struct kvec *kvec, size_t count, size_t seek)
 359 {
 360         iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
 361         return xs_sock_recvmsg(sock, msg, flags, seek);
 362 }
 363
 364 static ssize_t
 365 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 366                 struct bio_vec *bvec, unsigned long nr, size_t count,
 367                 size_t seek)
 368 {
 369         iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
 370         return xs_sock_recvmsg(sock, msg, flags, seek);
 371 }
 372
 373 static ssize_t
 374 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 375                 size_t count)
 376 {
 377         iov_iter_discard(&msg->msg_iter, READ, count);
 378         return sock_recvmsg(sock, msg, flags);
 379 }
 380
 381 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 382 static void
 383 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 384 {
 385         struct bvec_iter bi = {
 386                 .bi_size = count,
 387         };
 388         struct bio_vec bv;
 389
 390         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 391         for_each_bvec(bv, bvec, bi, bi)
 392                 flush_dcache_page(bv.bv_page);
 393 }
 394 #else
 395 static inline void
 396 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 397 {
 398 }
 399 #endif
 400
 401 static ssize_t
 402 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 403                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 404 {
 405         size_t want, seek_init = seek, offset = 0;
 406         ssize_t ret;
 407
 408         want = min_t(size_t, count, buf->head[0].iov_len);
 409         if (seek < want) {
 410                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 411                 if (ret <= 0)
 412                         goto sock_err;
 413                 offset += ret;
 414                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 415                         goto out;
 416                 if (ret != want)
 417                         goto out;
 418                 seek = 0;
 419         } else {
 420                 seek -= want;
 421                 offset += want;
 422         }
 423
 424         want = xs_alloc_sparse_pages(buf,
 425                         min_t(size_t, count - offset, buf->page_len),
 426                         GFP_KERNEL);
 427         if (seek < want) {
 428                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 429                                 xdr_buf_pagecount(buf),
 430                                 want + buf->page_base,
 431                                 seek + buf->page_base);
 432                 if (ret <= 0)
 433                         goto sock_err;
 434                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 435                 offset += ret - buf->page_base;
 436                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 437                         goto out;
 438                 if (ret != want)
 439                         goto out;
 440                 seek = 0;
 441         } else {
 442                 seek -= want;
 443                 offset += want;
 444         }
 445
 446         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 447         if (seek < want) {
 448                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 449                 if (ret <= 0)
 450                         goto sock_err;
 451                 offset += ret;
 452                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 453                         goto out;
 454                 if (ret != want)
 455                         goto out;
 456         } else if (offset < seek_init)
 457                 offset = seek_init;
 458         ret = -EMSGSIZE;
 459 out:
 460         *read = offset - seek_init;
 461         return ret;
 462 sock_err:
 463         offset += seek;
 464         goto out;
 465 }
 466
 467 static void
 468 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 469 {
 470         if (!transport->recv.copied) {
 471                 if (buf->head[0].iov_len >= transport->recv.offset)
 472                         memcpy(buf->head[0].iov_base,
 473                                         &transport->recv.xid,
 474                                         transport->recv.offset);
 475                 transport->recv.copied = transport->recv.offset;
 476         }
 477 }
 478
 479 static bool
 480 xs_read_stream_request_done(struct sock_xprt *transport)
 481 {
 482         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 483 }
 484
 485 static void
 486 xs_read_stream_check_eor(struct sock_xprt *transport,
 487                 struct msghdr *msg)
 488 {
 489         if (xs_read_stream_request_done(transport))
 490                 msg->msg_flags |= MSG_EOR;
 491 }
 492
 493 static ssize_t
 494 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 495                 int flags, struct rpc_rqst *req)
 496 {
 497         struct xdr_buf *buf = &req->rq_private_buf;
 498         size_t want, uninitialized_var(read);
 499         ssize_t uninitialized_var(ret);
 500
 501         xs_read_header(transport, buf);
 502
 503         want = transport->recv.len - transport->recv.offset;
 504         if (want != 0) {
 505                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 506                                 transport->recv.copied + want,
 507                                 transport->recv.copied,
 508                                 &read);
 509                 transport->recv.offset += read;
 510                 transport->recv.copied += read;
 511         }
 512
 513         if (transport->recv.offset == transport->recv.len)
 514                 xs_read_stream_check_eor(transport, msg);
 515
 516         if (want == 0)
 517                 return 0;
 518
 519         switch (ret) {
 520         default:
 521                 break;
 522         case -EFAULT:
 523         case -EMSGSIZE:
 524                 msg->msg_flags |= MSG_TRUNC;
 525                 return read;
 526         case 0:
 527                 return -ESHUTDOWN;
 528         }
 529         return ret < 0 ? ret : read;
 530 }
 531
 532 static size_t
 533 xs_read_stream_headersize(bool isfrag)
 534 {
 535         if (isfrag)
 536                 return sizeof(__be32);
 537         return 3 * sizeof(__be32);
 538 }
 539
 540 static ssize_t
 541 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 542                 int flags, size_t want, size_t seek)
 543 {
 544         struct kvec kvec = {
 545                 .iov_base = &transport->recv.fraghdr,
 546                 .iov_len = want,
 547         };
 548         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 549 }
 550
 551 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 552 static ssize_t
 553 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 554 {
 555         struct rpc_xprt *xprt = &transport->xprt;
 556         struct rpc_rqst *req;
 557         ssize_t ret;
 558
 559         /* Look up and lock the request corresponding to the given XID */
 560         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 561         if (!req) {
 562                 printk(KERN_WARNING "Callback slot table overflowed\n");
 563                 return -ESHUTDOWN;
 564         }
 565         if (transport->recv.copied && !req->rq_private_buf.len)
 566                 return -ESHUTDOWN;
 567
 568         ret = xs_read_stream_request(transport, msg, flags, req);
 569         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 570                 xprt_complete_bc_request(req, transport->recv.copied);
 571         else
 572                 req->rq_private_buf.len = transport->recv.copied;
 573
 574         return ret;
 575 }
 576 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 577 static ssize_t
 578 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 579 {
 580         return -ESHUTDOWN;
 581 }
 582 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 583
 584 static ssize_t
 585 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 586 {
 587         struct rpc_xprt *xprt = &transport->xprt;
 588         struct rpc_rqst *req;
 589         ssize_t ret = 0;
 590
 591         /* Look up and lock the request corresponding to the given XID */
 592         spin_lock(&xprt->queue_lock);
 593         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 594         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 595                 msg->msg_flags |= MSG_TRUNC;
 596                 goto out;
 597         }
 598         xprt_pin_rqst(req);
 599         spin_unlock(&xprt->queue_lock);
 600
 601         ret = xs_read_stream_request(transport, msg, flags, req);
 602
 603         spin_lock(&xprt->queue_lock);
 604         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 605                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 606         else
 607                 req->rq_private_buf.len = transport->recv.copied;
 608         xprt_unpin_rqst(req);
 609 out:
 610         spin_unlock(&xprt->queue_lock);
 611         return ret;
 612 }
 613
 614 static ssize_t
 615 xs_read_stream(struct sock_xprt *transport, int flags)
 616 {
 617         struct msghdr msg = { 0 };
 618         size_t want, read = 0;
 619         ssize_t ret = 0;
 620
 621         if (transport->recv.len == 0) {
 622                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 623                 ret = xs_read_stream_header(transport, &msg, flags, want,
 624                                 transport->recv.offset);
 625                 if (ret <= 0)
 626                         goto out_err;
 627                 transport->recv.offset = ret;
 628                 if (transport->recv.offset != want)
 629                         return transport->recv.offset;
 630                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 631                         RPC_FRAGMENT_SIZE_MASK;
 632                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 633                 read = ret;
 634         }
 635
 636         switch (be32_to_cpu(transport->recv.calldir)) {
 637         default:
 638                 msg.msg_flags |= MSG_TRUNC;
 639                 break;
 640         case RPC_CALL:
 641                 ret = xs_read_stream_call(transport, &msg, flags);
 642                 break;
 643         case RPC_REPLY:
 644                 ret = xs_read_stream_reply(transport, &msg, flags);
 645         }
 646         if (msg.msg_flags & MSG_TRUNC) {
 647                 transport->recv.calldir = cpu_to_be32(-1);
 648                 transport->recv.copied = -1;
 649         }
 650         if (ret < 0)
 651                 goto out_err;
 652         read += ret;
 653         if (transport->recv.offset < transport->recv.len) {
 654                 if (!(msg.msg_flags & MSG_TRUNC))
 655                         return read;
 656                 msg.msg_flags = 0;
 657                 ret = xs_read_discard(transport->sock, &msg, flags,
 658                                 transport->recv.len - transport->recv.offset);
 659                 if (ret <= 0)
 660                         goto out_err;
 661                 transport->recv.offset += ret;
 662                 read += ret;
 663                 if (transport->recv.offset != transport->recv.len)
 664                         return read;
 665         }
 666         if (xs_read_stream_request_done(transport)) {
 667                 trace_xs_stream_read_request(transport);
 668                 transport->recv.copied = 0;
 669         }
 670         transport->recv.offset = 0;
 671         transport->recv.len = 0;
 672         return read;
 673 out_err:
 674         return ret != 0 ? ret : -ESHUTDOWN;
 675 }
 676
 677 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 678 {
 679         return transport->sock->ops->poll(transport->file, transport->sock,
 680                         NULL);
 681 }
 682
 683 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 684 {
 685         __poll_t events = xs_poll_socket(transport);
 686
 687         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 688 }
 689
 690 static void xs_poll_check_readable(struct sock_xprt *transport)
 691 {
 692
 693         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 694         if (!xs_poll_socket_readable(transport))
 695                 return;
 696         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 697                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 698 }
 699
 700 static void xs_stream_data_receive(struct sock_xprt *transport)
 701 {
 702         size_t read = 0;
 703         ssize_t ret = 0;
 704
 705         mutex_lock(&transport->recv_mutex);
 706         if (transport->sock == NULL)
 707                 goto out;
 708         for (;;) {
 709                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 710                 if (ret < 0)
 711                         break;
 712                 read += ret;
 713                 cond_resched();
 714         }
 715         if (ret == -ESHUTDOWN)
 716                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 717         else
 718                 xs_poll_check_readable(transport);
 719 out:
 720         mutex_unlock(&transport->recv_mutex);
 721         trace_xs_stream_read_data(&transport->xprt, ret, read);
 722 }
 723
 724 static void xs_stream_data_receive_workfn(struct work_struct *work)
 725 {
 726         struct sock_xprt *transport =
 727                 container_of(work, struct sock_xprt, recv_worker);
 728         unsigned int pflags = memalloc_nofs_save();
 729
 730         xs_stream_data_receive(transport);
 731         memalloc_nofs_restore(pflags);
 732 }
 733
 734 static void
 735 xs_stream_reset_connect(struct sock_xprt *transport)
 736 {
 737         transport->recv.offset = 0;
 738         transport->recv.len = 0;
 739         transport->recv.copied = 0;
 740         transport->xmit.offset = 0;
 741 }
 742
 743 static void
 744 xs_stream_start_connect(struct sock_xprt *transport)
 745 {
 746         transport->xprt.stat.connect_count++;
 747         transport->xprt.stat.connect_start = jiffies;
 748 }
 749
 750 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 751
 752 static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
 753 {
 754         if (seek)
 755                 iov_iter_advance(&msg->msg_iter, seek);
 756         return sock_sendmsg(sock, msg);
 757 }
 758
 759 static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
 760 {
 761         iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
 762         return xs_sendmsg(sock, msg, seek);
 763 }
 764
 765 static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
 766 {
 767         int err;
 768
 769         err = xdr_alloc_bvec(xdr, GFP_KERNEL);
 770         if (err < 0)
 771                 return err;
 772
 773         iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
 774                         xdr_buf_pagecount(xdr),
 775                         xdr->page_len + xdr->page_base);
 776         return xs_sendmsg(sock, msg, base + xdr->page_base);
 777 }
 778
 779 #define xs_record_marker_len() sizeof(rpc_fraghdr)
 780
 781 /* Common case:
 782  *  - stream transport
 783  *  - sending from byte 0 of the message
 784  *  - the message is wholly contained in @xdr's head iovec
 785  */
 786 static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
 787                 rpc_fraghdr marker, struct kvec *vec, size_t base)
 788 {
 789         struct kvec iov[2] = {
 790                 [0] = {
 791                         .iov_base       = &marker,
 792                         .iov_len        = sizeof(marker)
 793                 },
 794                 [1] = *vec,
 795         };
 796         size_t len = iov[0].iov_len + iov[1].iov_len;
 797
 798         iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
 799         return xs_sendmsg(sock, msg, base);
 800 }
 801
 802 /**
 803  * xs_sendpages - write pages directly to a socket
 804  * @sock: socket to send on
 805  * @addr: UDP only -- address of destination
 806  * @addrlen: UDP only -- length of destination address
 807  * @xdr: buffer containing this request
 808  * @base: starting position in the buffer
 809  * @rm: stream record marker field
 810  * @sent_p: return the total number of bytes successfully queued for sending
 811  *
 812  */
 813 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
 814 {
 815         struct msghdr msg = {
 816                 .msg_name = addr,
 817                 .msg_namelen = addrlen,
 818                 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
 819         };
 820         unsigned int rmsize = rm ? sizeof(rm) : 0;
 821         unsigned int remainder = rmsize + xdr->len - base;
 822         unsigned int want;
 823         int err = 0;
 824
 825         if (unlikely(!sock))
 826                 return -ENOTSOCK;
 827
 828         want = xdr->head[0].iov_len + rmsize;
 829         if (base < want) {
 830                 unsigned int len = want - base;
 831                 remainder -= len;
 832                 if (remainder == 0)
 833                         msg.msg_flags &= ~MSG_MORE;
 834                 if (rmsize)
 835                         err = xs_send_rm_and_kvec(sock, &msg, rm,
 836                                         &xdr->head[0], base);
 837                 else
 838                         err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
 839                 if (remainder == 0 || err != len)
 840                         goto out;
 841                 *sent_p += err;
 842                 base = 0;
 843         } else
 844                 base -= want;
 845
 846         if (base < xdr->page_len) {
 847                 unsigned int len = xdr->page_len - base;
 848                 remainder -= len;
 849                 if (remainder == 0)
 850                         msg.msg_flags &= ~MSG_MORE;
 851                 err = xs_send_pagedata(sock, &msg, xdr, base);
 852                 if (remainder == 0 || err != len)
 853                         goto out;
 854                 *sent_p += err;
 855                 base = 0;
 856         } else
 857                 base -= xdr->page_len;
 858
 859         if (base >= xdr->tail[0].iov_len)
 860                 return 0;
 861         msg.msg_flags &= ~MSG_MORE;
 862         err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
 863 out:
 864         if (err > 0) {
 865                 *sent_p += err;
 866                 err = 0;
 867         }
 868         return err;
 869 }
 870
 871 /**
 872  * xs_nospace - handle transmit was incomplete
 873  * @req: pointer to RPC request
 874  *
 875  */
 876 static int xs_nospace(struct rpc_rqst *req)
 877 {
 878         struct rpc_xprt *xprt = req->rq_xprt;
 879         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 880         struct sock *sk = transport->inet;
 881         int ret = -EAGAIN;
 882
 883         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 884                         req->rq_task->tk_pid,
 885                         req->rq_slen - transport->xmit.offset,
 886                         req->rq_slen);
 887
 888         /* Protect against races with write_space */
 889         spin_lock(&xprt->transport_lock);
 890
 891         /* Don't race with disconnect */
 892         if (xprt_connected(xprt)) {
 893                 /* wait for more buffer space */
 894                 sk->sk_write_pending++;
 895                 xprt_wait_for_buffer_space(xprt);
 896         } else
 897                 ret = -ENOTCONN;
 898
 899         spin_unlock(&xprt->transport_lock);
 900
 901         /* Race breaker in case memory is freed before above code is called */
 902         if (ret == -EAGAIN) {
 903                 struct socket_wq *wq;
 904
 905                 rcu_read_lock();
 906                 wq = rcu_dereference(sk->sk_wq);
 907                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 908                 rcu_read_unlock();
 909
 910                 sk->sk_write_space(sk);
 911         }
 912         return ret;
 913 }
 914
 915 static void
 916 xs_stream_prepare_request(struct rpc_rqst *req)
 917 {
 918         xdr_free_bvec(&req->rq_rcv_buf);
 919         req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
 920 }
 921
 922 /*
 923  * Determine if the previous message in the stream was aborted before it
 924  * could complete transmission.
 925  */
 926 static bool
 927 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 928 {
 929         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 930 }
 931
 932 /*
 933  * Return the stream record marker field for a record of length < 2^31-1
 934  */
 935 static rpc_fraghdr
 936 xs_stream_record_marker(struct xdr_buf *xdr)
 937 {
 938         if (!xdr->len)
 939                 return 0;
 940         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 941 }
 942
 943 /**
 944  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 945  * @req: pointer to RPC request
 946  *
 947  * Return values:
 948  *        0:    The request has been sent
 949  *   EAGAIN:    The socket was blocked, please call again later to
 950  *              complete the request
 951  * ENOTCONN:    Caller needs to invoke connect logic then call again
 952  *    other:    Some other error occured, the request was not sent
 953  */
 954 static int xs_local_send_request(struct rpc_rqst *req)
 955 {
 956         struct rpc_xprt *xprt = req->rq_xprt;
 957         struct sock_xprt *transport =
 958                                 container_of(xprt, struct sock_xprt, xprt);
 959         struct xdr_buf *xdr = &req->rq_snd_buf;
 960         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 961         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 962         int status;
 963         int sent = 0;
 964
 965         /* Close the stream if the previous transmission was incomplete */
 966         if (xs_send_request_was_aborted(transport, req)) {
 967                 xs_close(xprt);
 968                 return -ENOTCONN;
 969         }
 970
 971         xs_pktdump("packet data:",
 972                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 973
 974         req->rq_xtime = ktime_get();
 975         status = xs_sendpages(transport->sock, NULL, 0, xdr,
 976                               transport->xmit.offset, rm, &sent);
 977         dprintk("RPC:       %s(%u) = %d\n",
 978                         __func__, xdr->len - transport->xmit.offset, status);
 979
 980         if (status == -EAGAIN && sock_writeable(transport->inet))
 981                 status = -ENOBUFS;
 982
 983         if (likely(sent > 0) || status == 0) {
 984                 transport->xmit.offset += sent;
 985                 req->rq_bytes_sent = transport->xmit.offset;
 986                 if (likely(req->rq_bytes_sent >= msglen)) {
 987                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 988                         transport->xmit.offset = 0;
 989                         return 0;
 990                 }
 991                 status = -EAGAIN;
 992         }
 993
 994         switch (status) {
 995         case -ENOBUFS:
 996                 break;
 997         case -EAGAIN:
 998                 status = xs_nospace(req);
 999                 break;
1000         default:
1001                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1002                         -status);
1003                 /* fall through */
1004         case -EPIPE:
1005                 xs_close(xprt);
1006                 status = -ENOTCONN;
1007         }
1008
1009         return status;
1010 }
1011
1012 /**
1013  * xs_udp_send_request - write an RPC request to a UDP socket
1014  * @req: pointer to RPC request
1015  *
1016  * Return values:
1017  *        0:    The request has been sent
1018  *   EAGAIN:    The socket was blocked, please call again later to
1019  *              complete the request
1020  * ENOTCONN:    Caller needs to invoke connect logic then call again
1021  *    other:    Some other error occurred, the request was not sent
1022  */
1023 static int xs_udp_send_request(struct rpc_rqst *req)
1024 {
1025         struct rpc_xprt *xprt = req->rq_xprt;
1026         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1027         struct xdr_buf *xdr = &req->rq_snd_buf;
1028         int sent = 0;
1029         int status;
1030
1031         xs_pktdump("packet data:",
1032                                 req->rq_svec->iov_base,
1033                                 req->rq_svec->iov_len);
1034
1035         if (!xprt_bound(xprt))
1036                 return -ENOTCONN;
1037
1038         if (!xprt_request_get_cong(xprt, req))
1039                 return -EBADSLT;
1040
1041         req->rq_xtime = ktime_get();
1042         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1043                               xdr, 0, 0, &sent);
1044
1045         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1046                         xdr->len, status);
1047
1048         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1049         if (status == -EPERM)
1050                 goto process_status;
1051
1052         if (status == -EAGAIN && sock_writeable(transport->inet))
1053                 status = -ENOBUFS;
1054
1055         if (sent > 0 || status == 0) {
1056                 req->rq_xmit_bytes_sent += sent;
1057                 if (sent >= req->rq_slen)
1058                         return 0;
1059                 /* Still some bytes left; set up for a retry later. */
1060                 status = -EAGAIN;
1061         }
1062
1063 process_status:
1064         switch (status) {
1065         case -ENOTSOCK:
1066                 status = -ENOTCONN;
1067                 /* Should we call xs_close() here? */
1068                 break;
1069         case -EAGAIN:
1070                 status = xs_nospace(req);
1071                 break;
1072         case -ENETUNREACH:
1073         case -ENOBUFS:
1074         case -EPIPE:
1075         case -ECONNREFUSED:
1076         case -EPERM:
1077                 /* When the server has died, an ICMP port unreachable message
1078                  * prompts ECONNREFUSED. */
1079                 break;
1080         default:
1081                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1082                         -status);
1083         }
1084
1085         return status;
1086 }
1087
1088 /**
1089  * xs_tcp_send_request - write an RPC request to a TCP socket
1090  * @req: pointer to RPC request
1091  *
1092  * Return values:
1093  *        0:    The request has been sent
1094  *   EAGAIN:    The socket was blocked, please call again later to
1095  *              complete the request
1096  * ENOTCONN:    Caller needs to invoke connect logic then call again
1097  *    other:    Some other error occurred, the request was not sent
1098  *
1099  * XXX: In the case of soft timeouts, should we eventually give up
1100  *      if sendmsg is not able to make progress?
1101  */
1102 static int xs_tcp_send_request(struct rpc_rqst *req)
1103 {
1104         struct rpc_xprt *xprt = req->rq_xprt;
1105         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1106         struct xdr_buf *xdr = &req->rq_snd_buf;
1107         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1108         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1109         bool vm_wait = false;
1110         int status;
1111         int sent;
1112
1113         /* Close the stream if the previous transmission was incomplete */
1114         if (xs_send_request_was_aborted(transport, req)) {
1115                 if (transport->sock != NULL)
1116                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1117                 return -ENOTCONN;
1118         }
1119
1120         xs_pktdump("packet data:",
1121                                 req->rq_svec->iov_base,
1122                                 req->rq_svec->iov_len);
1123
1124         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1125                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1126
1127         /* Continue transmitting the packet/record. We must be careful
1128          * to cope with writespace callbacks arriving _after_ we have
1129          * called sendmsg(). */
1130         req->rq_xtime = ktime_get();
1131         while (1) {
1132                 sent = 0;
1133                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1134                                       transport->xmit.offset, rm, &sent);
1135
1136                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1137                                 xdr->len - transport->xmit.offset, status);
1138
1139                 /* If we've sent the entire packet, immediately
1140                  * reset the count of bytes sent. */
1141                 transport->xmit.offset += sent;
1142                 req->rq_bytes_sent = transport->xmit.offset;
1143                 if (likely(req->rq_bytes_sent >= msglen)) {
1144                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1145                         transport->xmit.offset = 0;
1146                         return 0;
1147                 }
1148
1149                 WARN_ON_ONCE(sent == 0 && status == 0);
1150
1151                 if (status == -EAGAIN ) {
1152                         /*
1153                          * Return EAGAIN if we're sure we're hitting the
1154                          * socket send buffer limits.
1155                          */
1156                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1157                                 break;
1158                         /*
1159                          * Did we hit a memory allocation failure?
1160                          */
1161                         if (sent == 0) {
1162                                 status = -ENOBUFS;
1163                                 if (vm_wait)
1164                                         break;
1165                                 /* Retry, knowing now that we're below the
1166                                  * socket send buffer limit
1167                                  */
1168                                 vm_wait = true;
1169                         }
1170                         continue;
1171                 }
1172                 if (status < 0)
1173                         break;
1174                 vm_wait = false;
1175         }
1176
1177         switch (status) {
1178         case -ENOTSOCK:
1179                 status = -ENOTCONN;
1180                 /* Should we call xs_close() here? */
1181                 break;
1182         case -EAGAIN:
1183                 status = xs_nospace(req);
1184                 break;
1185         case -ECONNRESET:
1186         case -ECONNREFUSED:
1187         case -ENOTCONN:
1188         case -EADDRINUSE:
1189         case -ENOBUFS:
1190         case -EPIPE:
1191                 break;
1192         default:
1193                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1194                         -status);
1195         }
1196
1197         return status;
1198 }
1199
1200 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1201 {
1202         transport->old_data_ready = sk->sk_data_ready;
1203         transport->old_state_change = sk->sk_state_change;
1204         transport->old_write_space = sk->sk_write_space;
1205         transport->old_error_report = sk->sk_error_report;
1206 }
1207
1208 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1209 {
1210         sk->sk_data_ready = transport->old_data_ready;
1211         sk->sk_state_change = transport->old_state_change;
1212         sk->sk_write_space = transport->old_write_space;
1213         sk->sk_error_report = transport->old_error_report;
1214 }
1215
1216 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1217 {
1218         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1219
1220         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1221         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1222         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1223         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1224 }
1225
1226 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1227 {
1228         set_bit(nr, &transport->sock_state);
1229         queue_work(xprtiod_workqueue, &transport->error_worker);
1230 }
1231
1232 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1233 {
1234         smp_mb__before_atomic();
1235         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1236         clear_bit(XPRT_CLOSING, &xprt->state);
1237         xs_sock_reset_state_flags(xprt);
1238         smp_mb__after_atomic();
1239 }
1240
1241 /**
1242  * xs_error_report - callback to handle TCP socket state errors
1243  * @sk: socket
1244  *
1245  * Note: we don't call sock_error() since there may be a rpc_task
1246  * using the socket, and so we don't want to clear sk->sk_err.
1247  */
1248 static void xs_error_report(struct sock *sk)
1249 {
1250         struct sock_xprt *transport;
1251         struct rpc_xprt *xprt;
1252
1253         read_lock_bh(&sk->sk_callback_lock);
1254         if (!(xprt = xprt_from_sock(sk)))
1255                 goto out;
1256
1257         transport = container_of(xprt, struct sock_xprt, xprt);
1258         transport->xprt_err = -sk->sk_err;
1259         if (transport->xprt_err == 0)
1260                 goto out;
1261         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1262                         xprt, -transport->xprt_err);
1263         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1264
1265         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1266         smp_mb__before_atomic();
1267         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1268  out:
1269         read_unlock_bh(&sk->sk_callback_lock);
1270 }
1271
1272 static void xs_reset_transport(struct sock_xprt *transport)
1273 {
1274         struct socket *sock = transport->sock;
1275         struct sock *sk = transport->inet;
1276         struct rpc_xprt *xprt = &transport->xprt;
1277         struct file *filp = transport->file;
1278
1279         if (sk == NULL)
1280                 return;
1281
1282         if (atomic_read(&transport->xprt.swapper))
1283                 sk_clear_memalloc(sk);
1284
1285         kernel_sock_shutdown(sock, SHUT_RDWR);
1286
1287         mutex_lock(&transport->recv_mutex);
1288         write_lock_bh(&sk->sk_callback_lock);
1289         transport->inet = NULL;
1290         transport->sock = NULL;
1291         transport->file = NULL;
1292
1293         sk->sk_user_data = NULL;
1294
1295         xs_restore_old_callbacks(transport, sk);
1296         xprt_clear_connected(xprt);
1297         write_unlock_bh(&sk->sk_callback_lock);
1298         xs_sock_reset_connection_flags(xprt);
1299         /* Reset stream record info */
1300         xs_stream_reset_connect(transport);
1301         mutex_unlock(&transport->recv_mutex);
1302
1303         trace_rpc_socket_close(xprt, sock);
1304         fput(filp);
1305
1306         xprt_disconnect_done(xprt);
1307 }
1308
1309 /**
1310  * xs_close - close a socket
1311  * @xprt: transport
1312  *
1313  * This is used when all requests are complete; ie, no DRC state remains
1314  * on the server we want to save.
1315  *
1316  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1317  * xs_reset_transport() zeroing the socket from underneath a writer.
1318  */
1319 static void xs_close(struct rpc_xprt *xprt)
1320 {
1321         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1322
1323         dprintk("RPC:       xs_close xprt %p\n", xprt);
1324
1325         xs_reset_transport(transport);
1326         xprt->reestablish_timeout = 0;
1327 }
1328
1329 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1330 {
1331         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1332                 xprt);
1333         xprt_disconnect_done(xprt);
1334 }
1335
1336 static void xs_xprt_free(struct rpc_xprt *xprt)
1337 {
1338         xs_free_peer_addresses(xprt);
1339         xprt_free(xprt);
1340 }
1341
1342 /**
1343  * xs_destroy - prepare to shutdown a transport
1344  * @xprt: doomed transport
1345  *
1346  */
1347 static void xs_destroy(struct rpc_xprt *xprt)
1348 {
1349         struct sock_xprt *transport = container_of(xprt,
1350                         struct sock_xprt, xprt);
1351         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1352
1353         cancel_delayed_work_sync(&transport->connect_worker);
1354         xs_close(xprt);
1355         cancel_work_sync(&transport->recv_worker);
1356         cancel_work_sync(&transport->error_worker);
1357         xs_xprt_free(xprt);
1358         module_put(THIS_MODULE);
1359 }
1360
1361 /**
1362  * xs_udp_data_read_skb - receive callback for UDP sockets
1363  * @xprt: transport
1364  * @sk: socket
1365  * @skb: skbuff
1366  *
1367  */
1368 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1369                 struct sock *sk,
1370                 struct sk_buff *skb)
1371 {
1372         struct rpc_task *task;
1373         struct rpc_rqst *rovr;
1374         int repsize, copied;
1375         u32 _xid;
1376         __be32 *xp;
1377
1378         repsize = skb->len;
1379         if (repsize < 4) {
1380                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1381                 return;
1382         }
1383
1384         /* Copy the XID from the skb... */
1385         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1386         if (xp == NULL)
1387                 return;
1388
1389         /* Look up and lock the request corresponding to the given XID */
1390         spin_lock(&xprt->queue_lock);
1391         rovr = xprt_lookup_rqst(xprt, *xp);
1392         if (!rovr)
1393                 goto out_unlock;
1394         xprt_pin_rqst(rovr);
1395         xprt_update_rtt(rovr->rq_task);
1396         spin_unlock(&xprt->queue_lock);
1397         task = rovr->rq_task;
1398
1399         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1400                 copied = repsize;
1401
1402         /* Suck it into the iovec, verify checksum if not done by hw. */
1403         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1404                 spin_lock(&xprt->queue_lock);
1405                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1406                 goto out_unpin;
1407         }
1408
1409
1410         spin_lock(&xprt->transport_lock);
1411         xprt_adjust_cwnd(xprt, task, copied);
1412         spin_unlock(&xprt->transport_lock);
1413         spin_lock(&xprt->queue_lock);
1414         xprt_complete_rqst(task, copied);
1415         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1416 out_unpin:
1417         xprt_unpin_rqst(rovr);
1418  out_unlock:
1419         spin_unlock(&xprt->queue_lock);
1420 }
1421
1422 static void xs_udp_data_receive(struct sock_xprt *transport)
1423 {
1424         struct sk_buff *skb;
1425         struct sock *sk;
1426         int err;
1427
1428         mutex_lock(&transport->recv_mutex);
1429         sk = transport->inet;
1430         if (sk == NULL)
1431                 goto out;
1432         for (;;) {
1433                 skb = skb_recv_udp(sk, 0, 1, &err);
1434                 if (skb == NULL)
1435                         break;
1436                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1437                 consume_skb(skb);
1438                 cond_resched();
1439         }
1440         xs_poll_check_readable(transport);
1441 out:
1442         mutex_unlock(&transport->recv_mutex);
1443 }
1444
1445 static void xs_udp_data_receive_workfn(struct work_struct *work)
1446 {
1447         struct sock_xprt *transport =
1448                 container_of(work, struct sock_xprt, recv_worker);
1449         unsigned int pflags = memalloc_nofs_save();
1450
1451         xs_udp_data_receive(transport);
1452         memalloc_nofs_restore(pflags);
1453 }
1454
1455 /**
1456  * xs_data_ready - "data ready" callback for UDP sockets
1457  * @sk: socket with data to read
1458  *
1459  */
1460 static void xs_data_ready(struct sock *sk)
1461 {
1462         struct rpc_xprt *xprt;
1463
1464         read_lock_bh(&sk->sk_callback_lock);
1465         dprintk("RPC:       xs_data_ready...\n");
1466         xprt = xprt_from_sock(sk);
1467         if (xprt != NULL) {
1468                 struct sock_xprt *transport = container_of(xprt,
1469                                 struct sock_xprt, xprt);
1470                 transport->old_data_ready(sk);
1471                 /* Any data means we had a useful conversation, so
1472                  * then we don't need to delay the next reconnect
1473                  */
1474                 if (xprt->reestablish_timeout)
1475                         xprt->reestablish_timeout = 0;
1476                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1477                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1478         }
1479         read_unlock_bh(&sk->sk_callback_lock);
1480 }
1481
1482 /*
1483  * Helper function to force a TCP close if the server is sending
1484  * junk and/or it has put us in CLOSE_WAIT
1485  */
1486 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1487 {
1488         xprt_force_disconnect(xprt);
1489 }
1490
1491 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1492 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1493 {
1494         return PAGE_SIZE;
1495 }
1496 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1497
1498 /**
1499  * xs_tcp_state_change - callback to handle TCP socket state changes
1500  * @sk: socket whose state has changed
1501  *
1502  */
1503 static void xs_tcp_state_change(struct sock *sk)
1504 {
1505         struct rpc_xprt *xprt;
1506         struct sock_xprt *transport;
1507
1508         read_lock_bh(&sk->sk_callback_lock);
1509         if (!(xprt = xprt_from_sock(sk)))
1510                 goto out;
1511         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1512         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1513                         sk->sk_state, xprt_connected(xprt),
1514                         sock_flag(sk, SOCK_DEAD),
1515                         sock_flag(sk, SOCK_ZAPPED),
1516                         sk->sk_shutdown);
1517
1518         transport = container_of(xprt, struct sock_xprt, xprt);
1519         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1520         switch (sk->sk_state) {
1521         case TCP_ESTABLISHED:
1522                 if (!xprt_test_and_set_connected(xprt)) {
1523                         xprt->connect_cookie++;
1524                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1525                         xprt_clear_connecting(xprt);
1526
1527                         xprt->stat.connect_count++;
1528                         xprt->stat.connect_time += (long)jiffies -
1529                                                    xprt->stat.connect_start;
1530                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1531                 }
1532                 break;
1533         case TCP_FIN_WAIT1:
1534                 /* The client initiated a shutdown of the socket */
1535                 xprt->connect_cookie++;
1536                 xprt->reestablish_timeout = 0;
1537                 set_bit(XPRT_CLOSING, &xprt->state);
1538                 smp_mb__before_atomic();
1539                 clear_bit(XPRT_CONNECTED, &xprt->state);
1540                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1541                 smp_mb__after_atomic();
1542                 break;
1543         case TCP_CLOSE_WAIT:
1544                 /* The server initiated a shutdown of the socket */
1545                 xprt->connect_cookie++;
1546                 clear_bit(XPRT_CONNECTED, &xprt->state);
1547                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1548                 /* fall through */
1549         case TCP_CLOSING:
1550                 /*
1551                  * If the server closed down the connection, make sure that
1552                  * we back off before reconnecting
1553                  */
1554                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1555                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1556                 break;
1557         case TCP_LAST_ACK:
1558                 set_bit(XPRT_CLOSING, &xprt->state);
1559                 smp_mb__before_atomic();
1560                 clear_bit(XPRT_CONNECTED, &xprt->state);
1561                 smp_mb__after_atomic();
1562                 break;
1563         case TCP_CLOSE:
1564                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1565                                         &transport->sock_state))
1566                         xprt_clear_connecting(xprt);
1567                 clear_bit(XPRT_CLOSING, &xprt->state);
1568                 /* Trigger the socket release */
1569                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1570         }
1571  out:
1572         read_unlock_bh(&sk->sk_callback_lock);
1573 }
1574
1575 static void xs_write_space(struct sock *sk)
1576 {
1577         struct socket_wq *wq;
1578         struct sock_xprt *transport;
1579         struct rpc_xprt *xprt;
1580
1581         if (!sk->sk_socket)
1582                 return;
1583         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1584
1585         if (unlikely(!(xprt = xprt_from_sock(sk))))
1586                 return;
1587         transport = container_of(xprt, struct sock_xprt, xprt);
1588         rcu_read_lock();
1589         wq = rcu_dereference(sk->sk_wq);
1590         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1591                 goto out;
1592
1593         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1594         sk->sk_write_pending--;
1595 out:
1596         rcu_read_unlock();
1597 }
1598
1599 /**
1600  * xs_udp_write_space - callback invoked when socket buffer space
1601  *                             becomes available
1602  * @sk: socket whose state has changed
1603  *
1604  * Called when more output buffer space is available for this socket.
1605  * We try not to wake our writers until they can make "significant"
1606  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1607  * with a bunch of small requests.
1608  */
1609 static void xs_udp_write_space(struct sock *sk)
1610 {
1611         read_lock_bh(&sk->sk_callback_lock);
1612
1613         /* from net/core/sock.c:sock_def_write_space */
1614         if (sock_writeable(sk))
1615                 xs_write_space(sk);
1616
1617         read_unlock_bh(&sk->sk_callback_lock);
1618 }
1619
1620 /**
1621  * xs_tcp_write_space - callback invoked when socket buffer space
1622  *                             becomes available
1623  * @sk: socket whose state has changed
1624  *
1625  * Called when more output buffer space is available for this socket.
1626  * We try not to wake our writers until they can make "significant"
1627  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1628  * with a bunch of small requests.
1629  */
1630 static void xs_tcp_write_space(struct sock *sk)
1631 {
1632         read_lock_bh(&sk->sk_callback_lock);
1633
1634         /* from net/core/stream.c:sk_stream_write_space */
1635         if (sk_stream_is_writeable(sk))
1636                 xs_write_space(sk);
1637
1638         read_unlock_bh(&sk->sk_callback_lock);
1639 }
1640
1641 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1642 {
1643         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1644         struct sock *sk = transport->inet;
1645
1646         if (transport->rcvsize) {
1647                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1648                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1649         }
1650         if (transport->sndsize) {
1651                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1652                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1653                 sk->sk_write_space(sk);
1654         }
1655 }
1656
1657 /**
1658  * xs_udp_set_buffer_size - set send and receive limits
1659  * @xprt: generic transport
1660  * @sndsize: requested size of send buffer, in bytes
1661  * @rcvsize: requested size of receive buffer, in bytes
1662  *
1663  * Set socket send and receive buffer size limits.
1664  */
1665 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1666 {
1667         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1668
1669         transport->sndsize = 0;
1670         if (sndsize)
1671                 transport->sndsize = sndsize + 1024;
1672         transport->rcvsize = 0;
1673         if (rcvsize)
1674                 transport->rcvsize = rcvsize + 1024;
1675
1676         xs_udp_do_set_buffer_size(xprt);
1677 }
1678
1679 /**
1680  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1681  * @xprt: controlling transport
1682  * @task: task that timed out
1683  *
1684  * Adjust the congestion window after a retransmit timeout has occurred.
1685  */
1686 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1687 {
1688         spin_lock(&xprt->transport_lock);
1689         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1690         spin_unlock(&xprt->transport_lock);
1691 }
1692
1693 static int xs_get_random_port(void)
1694 {
1695         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1696         unsigned short range;
1697         unsigned short rand;
1698
1699         if (max < min)
1700                 return -EADDRINUSE;
1701         range = max - min + 1;
1702         rand = (unsigned short) prandom_u32() % range;
1703         return rand + min;
1704 }
1705
1706 /**
1707  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1708  * @sock: socket
1709  *
1710  * Note that this function has to be called on all sockets that share the
1711  * same port, and it must be called before binding.
1712  */
1713 static void xs_sock_set_reuseport(struct socket *sock)
1714 {
1715         int opt = 1;
1716
1717         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1718                         (char *)&opt, sizeof(opt));
1719 }
1720
1721 static unsigned short xs_sock_getport(struct socket *sock)
1722 {
1723         struct sockaddr_storage buf;
1724         unsigned short port = 0;
1725
1726         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1727                 goto out;
1728         switch (buf.ss_family) {
1729         case AF_INET6:
1730                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1731                 break;
1732         case AF_INET:
1733                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1734         }
1735 out:
1736         return port;
1737 }
1738
1739 /**
1740  * xs_set_port - reset the port number in the remote endpoint address
1741  * @xprt: generic transport
1742  * @port: new port number
1743  *
1744  */
1745 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1746 {
1747         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1748
1749         rpc_set_port(xs_addr(xprt), port);
1750         xs_update_peer_port(xprt);
1751 }
1752
1753 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1754 {
1755         if (transport->srcport == 0)
1756                 transport->srcport = xs_sock_getport(sock);
1757 }
1758
1759 static int xs_get_srcport(struct sock_xprt *transport)
1760 {
1761         int port = transport->srcport;
1762
1763         if (port == 0 && transport->xprt.resvport)
1764                 port = xs_get_random_port();
1765         return port;
1766 }
1767
1768 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1769 {
1770         if (transport->srcport != 0)
1771                 transport->srcport = 0;
1772         if (!transport->xprt.resvport)
1773                 return 0;
1774         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1775                 return xprt_max_resvport;
1776         return --port;
1777 }
1778 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1779 {
1780         struct sockaddr_storage myaddr;
1781         int err, nloop = 0;
1782         int port = xs_get_srcport(transport);
1783         unsigned short last;
1784
1785         /*
1786          * If we are asking for any ephemeral port (i.e. port == 0 &&
1787          * transport->xprt.resvport == 0), don't bind.  Let the local
1788          * port selection happen implicitly when the socket is used
1789          * (for example at connect time).
1790          *
1791          * This ensures that we can continue to establish TCP
1792          * connections even when all local ephemeral ports are already
1793          * a part of some TCP connection.  This makes no difference
1794          * for UDP sockets, but also doens't harm them.
1795          *
1796          * If we're asking for any reserved port (i.e. port == 0 &&
1797          * transport->xprt.resvport == 1) xs_get_srcport above will
1798          * ensure that port is non-zero and we will bind as needed.
1799          */
1800         if (port <= 0)
1801                 return port;
1802
1803         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1804         do {
1805                 rpc_set_port((struct sockaddr *)&myaddr, port);
1806                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1807                                 transport->xprt.addrlen);
1808                 if (err == 0) {
1809                         transport->srcport = port;
1810                         break;
1811                 }
1812                 last = port;
1813                 port = xs_next_srcport(transport, port);
1814                 if (port > last)
1815                         nloop++;
1816         } while (err == -EADDRINUSE && nloop != 2);
1817
1818         if (myaddr.ss_family == AF_INET)
1819                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1820                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1821                                 port, err ? "failed" : "ok", err);
1822         else
1823                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1824                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1825                                 port, err ? "failed" : "ok", err);
1826         return err;
1827 }
1828
1829 /*
1830  * We don't support autobind on AF_LOCAL sockets
1831  */
1832 static void xs_local_rpcbind(struct rpc_task *task)
1833 {
1834         xprt_set_bound(task->tk_xprt);
1835 }
1836
1837 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1838 {
1839 }
1840
1841 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1842 static struct lock_class_key xs_key[2];
1843 static struct lock_class_key xs_slock_key[2];
1844
1845 static inline void xs_reclassify_socketu(struct socket *sock)
1846 {
1847         struct sock *sk = sock->sk;
1848
1849         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1850                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1851 }
1852
1853 static inline void xs_reclassify_socket4(struct socket *sock)
1854 {
1855         struct sock *sk = sock->sk;
1856
1857         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1858                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1859 }
1860
1861 static inline void xs_reclassify_socket6(struct socket *sock)
1862 {
1863         struct sock *sk = sock->sk;
1864
1865         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1866                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1867 }
1868
1869 static inline void xs_reclassify_socket(int family, struct socket *sock)
1870 {
1871         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1872                 return;
1873
1874         switch (family) {
1875         case AF_LOCAL:
1876                 xs_reclassify_socketu(sock);
1877                 break;
1878         case AF_INET:
1879                 xs_reclassify_socket4(sock);
1880                 break;
1881         case AF_INET6:
1882                 xs_reclassify_socket6(sock);
1883                 break;
1884         }
1885 }
1886 #else
1887 static inline void xs_reclassify_socket(int family, struct socket *sock)
1888 {
1889 }
1890 #endif
1891
1892 static void xs_dummy_setup_socket(struct work_struct *work)
1893 {
1894 }
1895
1896 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1897                 struct sock_xprt *transport, int family, int type,
1898                 int protocol, bool reuseport)
1899 {
1900         struct file *filp;
1901         struct socket *sock;
1902         int err;
1903
1904         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1905         if (err < 0) {
1906                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1907                                 protocol, -err);
1908                 goto out;
1909         }
1910         xs_reclassify_socket(family, sock);
1911
1912         if (reuseport)
1913                 xs_sock_set_reuseport(sock);
1914
1915         err = xs_bind(transport, sock);
1916         if (err) {
1917                 sock_release(sock);
1918                 goto out;
1919         }
1920
1921         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1922         if (IS_ERR(filp))
1923                 return ERR_CAST(filp);
1924         transport->file = filp;
1925
1926         return sock;
1927 out:
1928         return ERR_PTR(err);
1929 }
1930
1931 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1932                                       struct socket *sock)
1933 {
1934         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935                                                                         xprt);
1936
1937         if (!transport->inet) {
1938                 struct sock *sk = sock->sk;
1939
1940                 write_lock_bh(&sk->sk_callback_lock);
1941
1942                 xs_save_old_callbacks(transport, sk);
1943
1944                 sk->sk_user_data = xprt;
1945                 sk->sk_data_ready = xs_data_ready;
1946                 sk->sk_write_space = xs_udp_write_space;
1947                 sock_set_flag(sk, SOCK_FASYNC);
1948                 sk->sk_error_report = xs_error_report;
1949
1950                 xprt_clear_connected(xprt);
1951
1952                 /* Reset to new socket */
1953                 transport->sock = sock;
1954                 transport->inet = sk;
1955
1956                 write_unlock_bh(&sk->sk_callback_lock);
1957         }
1958
1959         xs_stream_start_connect(transport);
1960
1961         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1962 }
1963
1964 /**
1965  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1966  * @transport: socket transport to connect
1967  */
1968 static int xs_local_setup_socket(struct sock_xprt *transport)
1969 {
1970         struct rpc_xprt *xprt = &transport->xprt;
1971         struct file *filp;
1972         struct socket *sock;
1973         int status = -EIO;
1974
1975         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1976                                         SOCK_STREAM, 0, &sock, 1);
1977         if (status < 0) {
1978                 dprintk("RPC:       can't create AF_LOCAL "
1979                         "transport socket (%d).\n", -status);
1980                 goto out;
1981         }
1982         xs_reclassify_socket(AF_LOCAL, sock);
1983
1984         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1985         if (IS_ERR(filp)) {
1986                 status = PTR_ERR(filp);
1987                 goto out;
1988         }
1989         transport->file = filp;
1990
1991         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1992                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1993
1994         status = xs_local_finish_connecting(xprt, sock);
1995         trace_rpc_socket_connect(xprt, sock, status);
1996         switch (status) {
1997         case 0:
1998                 dprintk("RPC:       xprt %p connected to %s\n",
1999                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2000                 xprt->stat.connect_count++;
2001                 xprt->stat.connect_time += (long)jiffies -
2002                                            xprt->stat.connect_start;
2003                 xprt_set_connected(xprt);
2004         case -ENOBUFS:
2005                 break;
2006         case -ENOENT:
2007                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2008                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2009                 break;
2010         case -ECONNREFUSED:
2011                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2012                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2013                 break;
2014         default:
2015                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2016                                 __func__, -status,
2017                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2018         }
2019
2020 out:
2021         xprt_clear_connecting(xprt);
2022         xprt_wake_pending_tasks(xprt, status);
2023         return status;
2024 }
2025
2026 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2027 {
2028         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2029         int ret;
2030
2031          if (RPC_IS_ASYNC(task)) {
2032                 /*
2033                  * We want the AF_LOCAL connect to be resolved in the
2034                  * filesystem namespace of the process making the rpc
2035                  * call.  Thus we connect synchronously.
2036                  *
2037                  * If we want to support asynchronous AF_LOCAL calls,
2038                  * we'll need to figure out how to pass a namespace to
2039                  * connect.
2040                  */
2041                 task->tk_rpc_status = -ENOTCONN;
2042                 rpc_exit(task, -ENOTCONN);
2043                 return;
2044         }
2045         ret = xs_local_setup_socket(transport);
2046         if (ret && !RPC_IS_SOFTCONN(task))
2047                 msleep_interruptible(15000);
2048 }
2049
2050 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2051 /*
2052  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2053  * know that we have exclusive access to the socket), to guard against
2054  * races with xs_reset_transport.
2055  */
2056 static void xs_set_memalloc(struct rpc_xprt *xprt)
2057 {
2058         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2059                         xprt);
2060
2061         /*
2062          * If there's no sock, then we have nothing to set. The
2063          * reconnecting process will get it for us.
2064          */
2065         if (!transport->inet)
2066                 return;
2067         if (atomic_read(&xprt->swapper))
2068                 sk_set_memalloc(transport->inet);
2069 }
2070
2071 /**
2072  * xs_enable_swap - Tag this transport as being used for swap.
2073  * @xprt: transport to tag
2074  *
2075  * Take a reference to this transport on behalf of the rpc_clnt, and
2076  * optionally mark it for swapping if it wasn't already.
2077  */
2078 static int
2079 xs_enable_swap(struct rpc_xprt *xprt)
2080 {
2081         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2082
2083         if (atomic_inc_return(&xprt->swapper) != 1)
2084                 return 0;
2085         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2086                 return -ERESTARTSYS;
2087         if (xs->inet)
2088                 sk_set_memalloc(xs->inet);
2089         xprt_release_xprt(xprt, NULL);
2090         return 0;
2091 }
2092
2093 /**
2094  * xs_disable_swap - Untag this transport as being used for swap.
2095  * @xprt: transport to tag
2096  *
2097  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2098  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2099  */
2100 static void
2101 xs_disable_swap(struct rpc_xprt *xprt)
2102 {
2103         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2104
2105         if (!atomic_dec_and_test(&xprt->swapper))
2106                 return;
2107         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2108                 return;
2109         if (xs->inet)
2110                 sk_clear_memalloc(xs->inet);
2111         xprt_release_xprt(xprt, NULL);
2112 }
2113 #else
2114 static void xs_set_memalloc(struct rpc_xprt *xprt)
2115 {
2116 }
2117
2118 static int
2119 xs_enable_swap(struct rpc_xprt *xprt)
2120 {
2121         return -EINVAL;
2122 }
2123
2124 static void
2125 xs_disable_swap(struct rpc_xprt *xprt)
2126 {
2127 }
2128 #endif
2129
2130 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2131 {
2132         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2133
2134         if (!transport->inet) {
2135                 struct sock *sk = sock->sk;
2136
2137                 write_lock_bh(&sk->sk_callback_lock);
2138
2139                 xs_save_old_callbacks(transport, sk);
2140
2141                 sk->sk_user_data = xprt;
2142                 sk->sk_data_ready = xs_data_ready;
2143                 sk->sk_write_space = xs_udp_write_space;
2144                 sock_set_flag(sk, SOCK_FASYNC);
2145
2146                 xprt_set_connected(xprt);
2147
2148                 /* Reset to new socket */
2149                 transport->sock = sock;
2150                 transport->inet = sk;
2151
2152                 xs_set_memalloc(xprt);
2153
2154                 write_unlock_bh(&sk->sk_callback_lock);
2155         }
2156         xs_udp_do_set_buffer_size(xprt);
2157
2158         xprt->stat.connect_start = jiffies;
2159 }
2160
2161 static void xs_udp_setup_socket(struct work_struct *work)
2162 {
2163         struct sock_xprt *transport =
2164                 container_of(work, struct sock_xprt, connect_worker.work);
2165         struct rpc_xprt *xprt = &transport->xprt;
2166         struct socket *sock;
2167         int status = -EIO;
2168
2169         sock = xs_create_sock(xprt, transport,
2170                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2171                         IPPROTO_UDP, false);
2172         if (IS_ERR(sock))
2173                 goto out;
2174
2175         dprintk("RPC:       worker connecting xprt %p via %s to "
2176                                 "%s (port %s)\n", xprt,
2177                         xprt->address_strings[RPC_DISPLAY_PROTO],
2178                         xprt->address_strings[RPC_DISPLAY_ADDR],
2179                         xprt->address_strings[RPC_DISPLAY_PORT]);
2180
2181         xs_udp_finish_connecting(xprt, sock);
2182         trace_rpc_socket_connect(xprt, sock, 0);
2183         status = 0;
2184 out:
2185         xprt_clear_connecting(xprt);
2186         xprt_unlock_connect(xprt, transport);
2187         xprt_wake_pending_tasks(xprt, status);
2188 }
2189
2190 /**
2191  * xs_tcp_shutdown - gracefully shut down a TCP socket
2192  * @xprt: transport
2193  *
2194  * Initiates a graceful shutdown of the TCP socket by calling the
2195  * equivalent of shutdown(SHUT_RDWR);
2196  */
2197 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2198 {
2199         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2200         struct socket *sock = transport->sock;
2201         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2202
2203         if (sock == NULL)
2204                 return;
2205         switch (skst) {
2206         default:
2207                 kernel_sock_shutdown(sock, SHUT_RDWR);
2208                 trace_rpc_socket_shutdown(xprt, sock);
2209                 break;
2210         case TCP_CLOSE:
2211         case TCP_TIME_WAIT:
2212                 xs_reset_transport(transport);
2213         }
2214 }
2215
2216 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2217                 struct socket *sock)
2218 {
2219         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2220         unsigned int keepidle;
2221         unsigned int keepcnt;
2222         unsigned int opt_on = 1;
2223         unsigned int timeo;
2224
2225         spin_lock(&xprt->transport_lock);
2226         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2227         keepcnt = xprt->timeout->to_retries + 1;
2228         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2229                 (xprt->timeout->to_retries + 1);
2230         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2231         spin_unlock(&xprt->transport_lock);
2232
2233         /* TCP Keepalive options */
2234         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2235                         (char *)&opt_on, sizeof(opt_on));
2236         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2237                         (char *)&keepidle, sizeof(keepidle));
2238         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2239                         (char *)&keepidle, sizeof(keepidle));
2240         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2241                         (char *)&keepcnt, sizeof(keepcnt));
2242
2243         /* TCP user timeout (see RFC5482) */
2244         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2245                         (char *)&timeo, sizeof(timeo));
2246 }
2247
2248 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2249                 unsigned long connect_timeout,
2250                 unsigned long reconnect_timeout)
2251 {
2252         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2253         struct rpc_timeout to;
2254         unsigned long initval;
2255
2256         spin_lock(&xprt->transport_lock);
2257         if (reconnect_timeout < xprt->max_reconnect_timeout)
2258                 xprt->max_reconnect_timeout = reconnect_timeout;
2259         if (connect_timeout < xprt->connect_timeout) {
2260                 memcpy(&to, xprt->timeout, sizeof(to));
2261                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2262                 /* Arbitrary lower limit */
2263                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2264                         initval = XS_TCP_INIT_REEST_TO << 1;
2265                 to.to_initval = initval;
2266                 to.to_maxval = initval;
2267                 memcpy(&transport->tcp_timeout, &to,
2268                                 sizeof(transport->tcp_timeout));
2269                 xprt->timeout = &transport->tcp_timeout;
2270                 xprt->connect_timeout = connect_timeout;
2271         }
2272         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2273         spin_unlock(&xprt->transport_lock);
2274 }
2275
2276 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2277 {
2278         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2279         int ret = -ENOTCONN;
2280
2281         if (!transport->inet) {
2282                 struct sock *sk = sock->sk;
2283                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2284
2285                 /* Avoid temporary address, they are bad for long-lived
2286                  * connections such as NFS mounts.
2287                  * RFC4941, section 3.6 suggests that:
2288                  *    Individual applications, which have specific
2289                  *    knowledge about the normal duration of connections,
2290                  *    MAY override this as appropriate.
2291                  */
2292                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2293                                 (char *)&addr_pref, sizeof(addr_pref));
2294
2295                 xs_tcp_set_socket_timeouts(xprt, sock);
2296
2297                 write_lock_bh(&sk->sk_callback_lock);
2298
2299                 xs_save_old_callbacks(transport, sk);
2300
2301                 sk->sk_user_data = xprt;
2302                 sk->sk_data_ready = xs_data_ready;
2303                 sk->sk_state_change = xs_tcp_state_change;
2304                 sk->sk_write_space = xs_tcp_write_space;
2305                 sock_set_flag(sk, SOCK_FASYNC);
2306                 sk->sk_error_report = xs_error_report;
2307
2308                 /* socket options */
2309                 sock_reset_flag(sk, SOCK_LINGER);
2310                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2311
2312                 xprt_clear_connected(xprt);
2313
2314                 /* Reset to new socket */
2315                 transport->sock = sock;
2316                 transport->inet = sk;
2317
2318                 write_unlock_bh(&sk->sk_callback_lock);
2319         }
2320
2321         if (!xprt_bound(xprt))
2322                 goto out;
2323
2324         xs_set_memalloc(xprt);
2325
2326         xs_stream_start_connect(transport);
2327
2328         /* Tell the socket layer to start connecting... */
2329         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2330         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2331         switch (ret) {
2332         case 0:
2333                 xs_set_srcport(transport, sock);
2334                 /* fall through */
2335         case -EINPROGRESS:
2336                 /* SYN_SENT! */
2337                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2338                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2339                 break;
2340         case -EADDRNOTAVAIL:
2341                 /* Source port number is unavailable. Try a new one! */
2342                 transport->srcport = 0;
2343         }
2344 out:
2345         return ret;
2346 }
2347
2348 /**
2349  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2350  * @work: queued work item
2351  *
2352  * Invoked by a work queue tasklet.
2353  */
2354 static void xs_tcp_setup_socket(struct work_struct *work)
2355 {
2356         struct sock_xprt *transport =
2357                 container_of(work, struct sock_xprt, connect_worker.work);
2358         struct socket *sock = transport->sock;
2359         struct rpc_xprt *xprt = &transport->xprt;
2360         int status = -EIO;
2361
2362         if (!sock) {
2363                 sock = xs_create_sock(xprt, transport,
2364                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2365                                 IPPROTO_TCP, true);
2366                 if (IS_ERR(sock)) {
2367                         status = PTR_ERR(sock);
2368                         goto out;
2369                 }
2370         }
2371
2372         dprintk("RPC:       worker connecting xprt %p via %s to "
2373                                 "%s (port %s)\n", xprt,
2374                         xprt->address_strings[RPC_DISPLAY_PROTO],
2375                         xprt->address_strings[RPC_DISPLAY_ADDR],
2376                         xprt->address_strings[RPC_DISPLAY_PORT]);
2377
2378         status = xs_tcp_finish_connecting(xprt, sock);
2379         trace_rpc_socket_connect(xprt, sock, status);
2380         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2381                         xprt, -status, xprt_connected(xprt),
2382                         sock->sk->sk_state);
2383         switch (status) {
2384         default:
2385                 printk("%s: connect returned unhandled error %d\n",
2386                         __func__, status);
2387                 /* fall through */
2388         case -EADDRNOTAVAIL:
2389                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2390                  * and retry
2391                  */
2392                 xs_tcp_force_close(xprt);
2393                 break;
2394         case 0:
2395         case -EINPROGRESS:
2396         case -EALREADY:
2397                 xprt_unlock_connect(xprt, transport);
2398                 return;
2399         case -EINVAL:
2400                 /* Happens, for instance, if the user specified a link
2401                  * local IPv6 address without a scope-id.
2402                  */
2403         case -ECONNREFUSED:
2404         case -ECONNRESET:
2405         case -ENETDOWN:
2406         case -ENETUNREACH:
2407         case -EHOSTUNREACH:
2408         case -EADDRINUSE:
2409         case -ENOBUFS:
2410                 /*
2411                  * xs_tcp_force_close() wakes tasks with -EIO.
2412                  * We need to wake them first to ensure the
2413                  * correct error code.
2414                  */
2415                 xprt_wake_pending_tasks(xprt, status);
2416                 xs_tcp_force_close(xprt);
2417                 goto out;
2418         }
2419         status = -EAGAIN;
2420 out:
2421         xprt_clear_connecting(xprt);
2422         xprt_unlock_connect(xprt, transport);
2423         xprt_wake_pending_tasks(xprt, status);
2424 }
2425
2426 /**
2427  * xs_connect - connect a socket to a remote endpoint
2428  * @xprt: pointer to transport structure
2429  * @task: address of RPC task that manages state of connect request
2430  *
2431  * TCP: If the remote end dropped the connection, delay reconnecting.
2432  *
2433  * UDP socket connects are synchronous, but we use a work queue anyway
2434  * to guarantee that even unprivileged user processes can set up a
2435  * socket on a privileged port.
2436  *
2437  * If a UDP socket connect fails, the delay behavior here prevents
2438  * retry floods (hard mounts).
2439  */
2440 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2441 {
2442         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2443         unsigned long delay = 0;
2444
2445         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2446
2447         if (transport->sock != NULL) {
2448                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2449                                 "seconds\n",
2450                                 xprt, xprt->reestablish_timeout / HZ);
2451
2452                 /* Start by resetting any existing state */
2453                 xs_reset_transport(transport);
2454
2455                 delay = xprt_reconnect_delay(xprt);
2456                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2457
2458         } else
2459                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2460
2461         queue_delayed_work(xprtiod_workqueue,
2462                         &transport->connect_worker,
2463                         delay);
2464 }
2465
2466 static void xs_wake_disconnect(struct sock_xprt *transport)
2467 {
2468         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2469                 xs_tcp_force_close(&transport->xprt);
2470 }
2471
2472 static void xs_wake_write(struct sock_xprt *transport)
2473 {
2474         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2475                 xprt_write_space(&transport->xprt);
2476 }
2477
2478 static void xs_wake_error(struct sock_xprt *transport)
2479 {
2480         int sockerr;
2481
2482         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2483                 return;
2484         mutex_lock(&transport->recv_mutex);
2485         if (transport->sock == NULL)
2486                 goto out;
2487         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2488                 goto out;
2489         sockerr = xchg(&transport->xprt_err, 0);
2490         if (sockerr < 0)
2491                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2492 out:
2493         mutex_unlock(&transport->recv_mutex);
2494 }
2495
2496 static void xs_wake_pending(struct sock_xprt *transport)
2497 {
2498         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2499                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2500 }
2501
2502 static void xs_error_handle(struct work_struct *work)
2503 {
2504         struct sock_xprt *transport = container_of(work,
2505                         struct sock_xprt, error_worker);
2506
2507         xs_wake_disconnect(transport);
2508         xs_wake_write(transport);
2509         xs_wake_error(transport);
2510         xs_wake_pending(transport);
2511 }
2512
2513 /**
2514  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2515  * @xprt: rpc_xprt struct containing statistics
2516  * @seq: output file
2517  *
2518  */
2519 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2520 {
2521         long idle_time = 0;
2522
2523         if (xprt_connected(xprt))
2524                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2525
2526         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2527                         "%llu %llu %lu %llu %llu\n",
2528                         xprt->stat.bind_count,
2529                         xprt->stat.connect_count,
2530                         xprt->stat.connect_time / HZ,
2531                         idle_time,
2532                         xprt->stat.sends,
2533                         xprt->stat.recvs,
2534                         xprt->stat.bad_xids,
2535                         xprt->stat.req_u,
2536                         xprt->stat.bklog_u,
2537                         xprt->stat.max_slots,
2538                         xprt->stat.sending_u,
2539                         xprt->stat.pending_u);
2540 }
2541
2542 /**
2543  * xs_udp_print_stats - display UDP socket-specifc stats
2544  * @xprt: rpc_xprt struct containing statistics
2545  * @seq: output file
2546  *
2547  */
2548 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2549 {
2550         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2551
2552         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2553                         "%lu %llu %llu\n",
2554                         transport->srcport,
2555                         xprt->stat.bind_count,
2556                         xprt->stat.sends,
2557                         xprt->stat.recvs,
2558                         xprt->stat.bad_xids,
2559                         xprt->stat.req_u,
2560                         xprt->stat.bklog_u,
2561                         xprt->stat.max_slots,
2562                         xprt->stat.sending_u,
2563                         xprt->stat.pending_u);
2564 }
2565
2566 /**
2567  * xs_tcp_print_stats - display TCP socket-specifc stats
2568  * @xprt: rpc_xprt struct containing statistics
2569  * @seq: output file
2570  *
2571  */
2572 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2573 {
2574         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2575         long idle_time = 0;
2576
2577         if (xprt_connected(xprt))
2578                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2579
2580         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2581                         "%llu %llu %lu %llu %llu\n",
2582                         transport->srcport,
2583                         xprt->stat.bind_count,
2584                         xprt->stat.connect_count,
2585                         xprt->stat.connect_time / HZ,
2586                         idle_time,
2587                         xprt->stat.sends,
2588                         xprt->stat.recvs,
2589                         xprt->stat.bad_xids,
2590                         xprt->stat.req_u,
2591                         xprt->stat.bklog_u,
2592                         xprt->stat.max_slots,
2593                         xprt->stat.sending_u,
2594                         xprt->stat.pending_u);
2595 }
2596
2597 /*
2598  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2599  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2600  * to use the server side send routines.
2601  */
2602 static int bc_malloc(struct rpc_task *task)
2603 {
2604         struct rpc_rqst *rqst = task->tk_rqstp;
2605         size_t size = rqst->rq_callsize;
2606         struct page *page;
2607         struct rpc_buffer *buf;
2608
2609         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2610                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2611                           size);
2612                 return -EINVAL;
2613         }
2614
2615         page = alloc_page(GFP_KERNEL);
2616         if (!page)
2617                 return -ENOMEM;
2618
2619         buf = page_address(page);
2620         buf->len = PAGE_SIZE;
2621
2622         rqst->rq_buffer = buf->data;
2623         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2624         return 0;
2625 }
2626
2627 /*
2628  * Free the space allocated in the bc_alloc routine
2629  */
2630 static void bc_free(struct rpc_task *task)
2631 {
2632         void *buffer = task->tk_rqstp->rq_buffer;
2633         struct rpc_buffer *buf;
2634
2635         buf = container_of(buffer, struct rpc_buffer, data);
2636         free_page((unsigned long)buf);
2637 }
2638
2639 /*
2640  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2641  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2642  */
2643 static int bc_sendto(struct rpc_rqst *req)
2644 {
2645         int len;
2646         struct xdr_buf *xbufp = &req->rq_snd_buf;
2647         struct sock_xprt *transport =
2648                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2649         unsigned long headoff;
2650         unsigned long tailoff;
2651         struct page *tailpage;
2652         struct msghdr msg = {
2653                 .msg_flags      = MSG_MORE
2654         };
2655         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2656                                          (u32)xbufp->len);
2657         struct kvec iov = {
2658                 .iov_base       = &marker,
2659                 .iov_len        = sizeof(marker),
2660         };
2661
2662         len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2663         if (len != iov.iov_len)
2664                 return -EAGAIN;
2665
2666         tailpage = NULL;
2667         if (xbufp->tail[0].iov_len)
2668                 tailpage = virt_to_page(xbufp->tail[0].iov_base);
2669         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2670         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2671         len = svc_send_common(transport->sock, xbufp,
2672                               virt_to_page(xbufp->head[0].iov_base), headoff,
2673                               tailpage, tailoff);
2674         if (len != xbufp->len)
2675                 return -EAGAIN;
2676         return len;
2677 }
2678
2679 /*
2680  * The send routine. Borrows from svc_send
2681  */
2682 static int bc_send_request(struct rpc_rqst *req)
2683 {
2684         struct svc_xprt *xprt;
2685         int len;
2686
2687         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2688         /*
2689          * Get the server socket associated with this callback xprt
2690          */
2691         xprt = req->rq_xprt->bc_xprt;
2692
2693         /*
2694          * Grab the mutex to serialize data as the connection is shared
2695          * with the fore channel
2696          */
2697         mutex_lock(&xprt->xpt_mutex);
2698         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2699                 len = -ENOTCONN;
2700         else
2701                 len = bc_sendto(req);
2702         mutex_unlock(&xprt->xpt_mutex);
2703
2704         if (len > 0)
2705                 len = 0;
2706
2707         return len;
2708 }
2709
2710 /*
2711  * The close routine. Since this is client initiated, we do nothing
2712  */
2713
2714 static void bc_close(struct rpc_xprt *xprt)
2715 {
2716 }
2717
2718 /*
2719  * The xprt destroy routine. Again, because this connection is client
2720  * initiated, we do nothing
2721  */
2722
2723 static void bc_destroy(struct rpc_xprt *xprt)
2724 {
2725         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2726
2727         xs_xprt_free(xprt);
2728         module_put(THIS_MODULE);
2729 }
2730
2731 static const struct rpc_xprt_ops xs_local_ops = {
2732         .reserve_xprt           = xprt_reserve_xprt,
2733         .release_xprt           = xprt_release_xprt,
2734         .alloc_slot             = xprt_alloc_slot,
2735         .free_slot              = xprt_free_slot,
2736         .rpcbind                = xs_local_rpcbind,
2737         .set_port               = xs_local_set_port,
2738         .connect                = xs_local_connect,
2739         .buf_alloc              = rpc_malloc,
2740         .buf_free               = rpc_free,
2741         .prepare_request        = xs_stream_prepare_request,
2742         .send_request           = xs_local_send_request,
2743         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2744         .close                  = xs_close,
2745         .destroy                = xs_destroy,
2746         .print_stats            = xs_local_print_stats,
2747         .enable_swap            = xs_enable_swap,
2748         .disable_swap           = xs_disable_swap,
2749 };
2750
2751 static const struct rpc_xprt_ops xs_udp_ops = {
2752         .set_buffer_size        = xs_udp_set_buffer_size,
2753         .reserve_xprt           = xprt_reserve_xprt_cong,
2754         .release_xprt           = xprt_release_xprt_cong,
2755         .alloc_slot             = xprt_alloc_slot,
2756         .free_slot              = xprt_free_slot,
2757         .rpcbind                = rpcb_getport_async,
2758         .set_port               = xs_set_port,
2759         .connect                = xs_connect,
2760         .buf_alloc              = rpc_malloc,
2761         .buf_free               = rpc_free,
2762         .send_request           = xs_udp_send_request,
2763         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2764         .timer                  = xs_udp_timer,
2765         .release_request        = xprt_release_rqst_cong,
2766         .close                  = xs_close,
2767         .destroy                = xs_destroy,
2768         .print_stats            = xs_udp_print_stats,
2769         .enable_swap            = xs_enable_swap,
2770         .disable_swap           = xs_disable_swap,
2771         .inject_disconnect      = xs_inject_disconnect,
2772 };
2773
2774 static const struct rpc_xprt_ops xs_tcp_ops = {
2775         .reserve_xprt           = xprt_reserve_xprt,
2776         .release_xprt           = xprt_release_xprt,
2777         .alloc_slot             = xprt_alloc_slot,
2778         .free_slot              = xprt_free_slot,
2779         .rpcbind                = rpcb_getport_async,
2780         .set_port               = xs_set_port,
2781         .connect                = xs_connect,
2782         .buf_alloc              = rpc_malloc,
2783         .buf_free               = rpc_free,
2784         .prepare_request        = xs_stream_prepare_request,
2785         .send_request           = xs_tcp_send_request,
2786         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2787         .close                  = xs_tcp_shutdown,
2788         .destroy                = xs_destroy,
2789         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2790         .print_stats            = xs_tcp_print_stats,
2791         .enable_swap            = xs_enable_swap,
2792         .disable_swap           = xs_disable_swap,
2793         .inject_disconnect      = xs_inject_disconnect,
2794 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2795         .bc_setup               = xprt_setup_bc,
2796         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2797         .bc_num_slots           = xprt_bc_max_slots,
2798         .bc_free_rqst           = xprt_free_bc_rqst,
2799         .bc_destroy             = xprt_destroy_bc,
2800 #endif
2801 };
2802
2803 /*
2804  * The rpc_xprt_ops for the server backchannel
2805  */
2806
2807 static const struct rpc_xprt_ops bc_tcp_ops = {
2808         .reserve_xprt           = xprt_reserve_xprt,
2809         .release_xprt           = xprt_release_xprt,
2810         .alloc_slot             = xprt_alloc_slot,
2811         .free_slot              = xprt_free_slot,
2812         .buf_alloc              = bc_malloc,
2813         .buf_free               = bc_free,
2814         .send_request           = bc_send_request,
2815         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2816         .close                  = bc_close,
2817         .destroy                = bc_destroy,
2818         .print_stats            = xs_tcp_print_stats,
2819         .enable_swap            = xs_enable_swap,
2820         .disable_swap           = xs_disable_swap,
2821         .inject_disconnect      = xs_inject_disconnect,
2822 };
2823
2824 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2825 {
2826         static const struct sockaddr_in sin = {
2827                 .sin_family             = AF_INET,
2828                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2829         };
2830         static const struct sockaddr_in6 sin6 = {
2831                 .sin6_family            = AF_INET6,
2832                 .sin6_addr              = IN6ADDR_ANY_INIT,
2833         };
2834
2835         switch (family) {
2836         case AF_LOCAL:
2837                 break;
2838         case AF_INET:
2839                 memcpy(sap, &sin, sizeof(sin));
2840                 break;
2841         case AF_INET6:
2842                 memcpy(sap, &sin6, sizeof(sin6));
2843                 break;
2844         default:
2845                 dprintk("RPC:       %s: Bad address family\n", __func__);
2846                 return -EAFNOSUPPORT;
2847         }
2848         return 0;
2849 }
2850
2851 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2852                                       unsigned int slot_table_size,
2853                                       unsigned int max_slot_table_size)
2854 {
2855         struct rpc_xprt *xprt;
2856         struct sock_xprt *new;
2857
2858         if (args->addrlen > sizeof(xprt->addr)) {
2859                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2860                 return ERR_PTR(-EBADF);
2861         }
2862
2863         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2864                         max_slot_table_size);
2865         if (xprt == NULL) {
2866                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2867                                 "rpc_xprt\n");
2868                 return ERR_PTR(-ENOMEM);
2869         }
2870
2871         new = container_of(xprt, struct sock_xprt, xprt);
2872         mutex_init(&new->recv_mutex);
2873         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2874         xprt->addrlen = args->addrlen;
2875         if (args->srcaddr)
2876                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2877         else {
2878                 int err;
2879                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2880                                         (struct sockaddr *)&new->srcaddr);
2881                 if (err != 0) {
2882                         xprt_free(xprt);
2883                         return ERR_PTR(err);
2884                 }
2885         }
2886
2887         return xprt;
2888 }
2889
2890 static const struct rpc_timeout xs_local_default_timeout = {
2891         .to_initval = 10 * HZ,
2892         .to_maxval = 10 * HZ,
2893         .to_retries = 2,
2894 };
2895
2896 /**
2897  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2898  * @args: rpc transport creation arguments
2899  *
2900  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2901  */
2902 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2903 {
2904         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2905         struct sock_xprt *transport;
2906         struct rpc_xprt *xprt;
2907         struct rpc_xprt *ret;
2908
2909         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2910                         xprt_max_tcp_slot_table_entries);
2911         if (IS_ERR(xprt))
2912                 return xprt;
2913         transport = container_of(xprt, struct sock_xprt, xprt);
2914
2915         xprt->prot = 0;
2916         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2917
2918         xprt->bind_timeout = XS_BIND_TO;
2919         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2920         xprt->idle_timeout = XS_IDLE_DISC_TO;
2921
2922         xprt->ops = &xs_local_ops;
2923         xprt->timeout = &xs_local_default_timeout;
2924
2925         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2926         INIT_WORK(&transport->error_worker, xs_error_handle);
2927         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2928
2929         switch (sun->sun_family) {
2930         case AF_LOCAL:
2931                 if (sun->sun_path[0] != '/') {
2932                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2933                                         sun->sun_path);
2934                         ret = ERR_PTR(-EINVAL);
2935                         goto out_err;
2936                 }
2937                 xprt_set_bound(xprt);
2938                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2939                 ret = ERR_PTR(xs_local_setup_socket(transport));
2940                 if (ret)
2941                         goto out_err;
2942                 break;
2943         default:
2944                 ret = ERR_PTR(-EAFNOSUPPORT);
2945                 goto out_err;
2946         }
2947
2948         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2949                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2950
2951         if (try_module_get(THIS_MODULE))
2952                 return xprt;
2953         ret = ERR_PTR(-EINVAL);
2954 out_err:
2955         xs_xprt_free(xprt);
2956         return ret;
2957 }
2958
2959 static const struct rpc_timeout xs_udp_default_timeout = {
2960         .to_initval = 5 * HZ,
2961         .to_maxval = 30 * HZ,
2962         .to_increment = 5 * HZ,
2963         .to_retries = 5,
2964 };
2965
2966 /**
2967  * xs_setup_udp - Set up transport to use a UDP socket
2968  * @args: rpc transport creation arguments
2969  *
2970  */
2971 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2972 {
2973         struct sockaddr *addr = args->dstaddr;
2974         struct rpc_xprt *xprt;
2975         struct sock_xprt *transport;
2976         struct rpc_xprt *ret;
2977
2978         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2979                         xprt_udp_slot_table_entries);
2980         if (IS_ERR(xprt))
2981                 return xprt;
2982         transport = container_of(xprt, struct sock_xprt, xprt);
2983
2984         xprt->prot = IPPROTO_UDP;
2985         /* XXX: header size can vary due to auth type, IPv6, etc. */
2986         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2987
2988         xprt->bind_timeout = XS_BIND_TO;
2989         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2990         xprt->idle_timeout = XS_IDLE_DISC_TO;
2991
2992         xprt->ops = &xs_udp_ops;
2993
2994         xprt->timeout = &xs_udp_default_timeout;
2995
2996         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2997         INIT_WORK(&transport->error_worker, xs_error_handle);
2998         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2999
3000         switch (addr->sa_family) {
3001         case AF_INET:
3002                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3003                         xprt_set_bound(xprt);
3004
3005                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3006                 break;
3007         case AF_INET6:
3008                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3009                         xprt_set_bound(xprt);
3010
3011                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3012                 break;
3013         default:
3014                 ret = ERR_PTR(-EAFNOSUPPORT);
3015                 goto out_err;
3016         }
3017
3018         if (xprt_bound(xprt))
3019                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3020                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3021                                 xprt->address_strings[RPC_DISPLAY_PORT],
3022                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3023         else
3024                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3025                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3026                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028         if (try_module_get(THIS_MODULE))
3029                 return xprt;
3030         ret = ERR_PTR(-EINVAL);
3031 out_err:
3032         xs_xprt_free(xprt);
3033         return ret;
3034 }
3035
3036 static const struct rpc_timeout xs_tcp_default_timeout = {
3037         .to_initval = 60 * HZ,
3038         .to_maxval = 60 * HZ,
3039         .to_retries = 2,
3040 };
3041
3042 /**
3043  * xs_setup_tcp - Set up transport to use a TCP socket
3044  * @args: rpc transport creation arguments
3045  *
3046  */
3047 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3048 {
3049         struct sockaddr *addr = args->dstaddr;
3050         struct rpc_xprt *xprt;
3051         struct sock_xprt *transport;
3052         struct rpc_xprt *ret;
3053         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3054
3055         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3056                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3057
3058         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3059                         max_slot_table_size);
3060         if (IS_ERR(xprt))
3061                 return xprt;
3062         transport = container_of(xprt, struct sock_xprt, xprt);
3063
3064         xprt->prot = IPPROTO_TCP;
3065         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3066
3067         xprt->bind_timeout = XS_BIND_TO;
3068         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3069         xprt->idle_timeout = XS_IDLE_DISC_TO;
3070
3071         xprt->ops = &xs_tcp_ops;
3072         xprt->timeout = &xs_tcp_default_timeout;
3073
3074         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3075         xprt->connect_timeout = xprt->timeout->to_initval *
3076                 (xprt->timeout->to_retries + 1);
3077
3078         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3079         INIT_WORK(&transport->error_worker, xs_error_handle);
3080         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3081
3082         switch (addr->sa_family) {
3083         case AF_INET:
3084                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3085                         xprt_set_bound(xprt);
3086
3087                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3088                 break;
3089         case AF_INET6:
3090                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3091                         xprt_set_bound(xprt);
3092
3093                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3094                 break;
3095         default:
3096                 ret = ERR_PTR(-EAFNOSUPPORT);
3097                 goto out_err;
3098         }
3099
3100         if (xprt_bound(xprt))
3101                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3102                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3103                                 xprt->address_strings[RPC_DISPLAY_PORT],
3104                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3105         else
3106                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3107                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3108                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3109
3110         if (try_module_get(THIS_MODULE))
3111                 return xprt;
3112         ret = ERR_PTR(-EINVAL);
3113 out_err:
3114         xs_xprt_free(xprt);
3115         return ret;
3116 }
3117
3118 /**
3119  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3120  * @args: rpc transport creation arguments
3121  *
3122  */
3123 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3124 {
3125         struct sockaddr *addr = args->dstaddr;
3126         struct rpc_xprt *xprt;
3127         struct sock_xprt *transport;
3128         struct svc_sock *bc_sock;
3129         struct rpc_xprt *ret;
3130
3131         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3132                         xprt_tcp_slot_table_entries);
3133         if (IS_ERR(xprt))
3134                 return xprt;
3135         transport = container_of(xprt, struct sock_xprt, xprt);
3136
3137         xprt->prot = IPPROTO_TCP;
3138         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3139         xprt->timeout = &xs_tcp_default_timeout;
3140
3141         /* backchannel */
3142         xprt_set_bound(xprt);
3143         xprt->bind_timeout = 0;
3144         xprt->reestablish_timeout = 0;
3145         xprt->idle_timeout = 0;
3146
3147         xprt->ops = &bc_tcp_ops;
3148
3149         switch (addr->sa_family) {
3150         case AF_INET:
3151                 xs_format_peer_addresses(xprt, "tcp",
3152                                          RPCBIND_NETID_TCP);
3153                 break;
3154         case AF_INET6:
3155                 xs_format_peer_addresses(xprt, "tcp",
3156                                    RPCBIND_NETID_TCP6);
3157                 break;
3158         default:
3159                 ret = ERR_PTR(-EAFNOSUPPORT);
3160                 goto out_err;
3161         }
3162
3163         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3164                         xprt->address_strings[RPC_DISPLAY_ADDR],
3165                         xprt->address_strings[RPC_DISPLAY_PORT],
3166                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3167
3168         /*
3169          * Once we've associated a backchannel xprt with a connection,
3170          * we want to keep it around as long as the connection lasts,
3171          * in case we need to start using it for a backchannel again;
3172          * this reference won't be dropped until bc_xprt is destroyed.
3173          */
3174         xprt_get(xprt);
3175         args->bc_xprt->xpt_bc_xprt = xprt;
3176         xprt->bc_xprt = args->bc_xprt;
3177         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3178         transport->sock = bc_sock->sk_sock;
3179         transport->inet = bc_sock->sk_sk;
3180
3181         /*
3182          * Since we don't want connections for the backchannel, we set
3183          * the xprt status to connected
3184          */
3185         xprt_set_connected(xprt);
3186
3187         if (try_module_get(THIS_MODULE))
3188                 return xprt;
3189
3190         args->bc_xprt->xpt_bc_xprt = NULL;
3191         args->bc_xprt->xpt_bc_xps = NULL;
3192         xprt_put(xprt);
3193         ret = ERR_PTR(-EINVAL);
3194 out_err:
3195         xs_xprt_free(xprt);
3196         return ret;
3197 }
3198
3199 static struct xprt_class        xs_local_transport = {
3200         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3201         .name           = "named UNIX socket",
3202         .owner          = THIS_MODULE,
3203         .ident          = XPRT_TRANSPORT_LOCAL,
3204         .setup          = xs_setup_local,
3205 };
3206
3207 static struct xprt_class        xs_udp_transport = {
3208         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3209         .name           = "udp",
3210         .owner          = THIS_MODULE,
3211         .ident          = XPRT_TRANSPORT_UDP,
3212         .setup          = xs_setup_udp,
3213 };
3214
3215 static struct xprt_class        xs_tcp_transport = {
3216         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3217         .name           = "tcp",
3218         .owner          = THIS_MODULE,
3219         .ident          = XPRT_TRANSPORT_TCP,
3220         .setup          = xs_setup_tcp,
3221 };
3222
3223 static struct xprt_class        xs_bc_tcp_transport = {
3224         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3225         .name           = "tcp NFSv4.1 backchannel",
3226         .owner          = THIS_MODULE,
3227         .ident          = XPRT_TRANSPORT_BC_TCP,
3228         .setup          = xs_setup_bc_tcp,
3229 };
3230
3231 /**
3232  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3233  *
3234  */
3235 int init_socket_xprt(void)
3236 {
3237         if (!sunrpc_table_header)
3238                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3239
3240         xprt_register_transport(&xs_local_transport);
3241         xprt_register_transport(&xs_udp_transport);
3242         xprt_register_transport(&xs_tcp_transport);
3243         xprt_register_transport(&xs_bc_tcp_transport);
3244
3245         return 0;
3246 }
3247
3248 /**
3249  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3250  *
3251  */
3252 void cleanup_socket_xprt(void)
3253 {
3254         if (sunrpc_table_header) {
3255                 unregister_sysctl_table(sunrpc_table_header);
3256                 sunrpc_table_header = NULL;
3257         }
3258
3259         xprt_unregister_transport(&xs_local_transport);
3260         xprt_unregister_transport(&xs_udp_transport);
3261         xprt_unregister_transport(&xs_tcp_transport);
3262         xprt_unregister_transport(&xs_bc_tcp_transport);
3263 }
3264
3265 static int param_set_uint_minmax(const char *val,
3266                 const struct kernel_param *kp,
3267                 unsigned int min, unsigned int max)
3268 {
3269         unsigned int num;
3270         int ret;
3271
3272         if (!val)
3273                 return -EINVAL;
3274         ret = kstrtouint(val, 0, &num);
3275         if (ret)
3276                 return ret;
3277         if (num < min || num > max)
3278                 return -EINVAL;
3279         *((unsigned int *)kp->arg) = num;
3280         return 0;
3281 }
3282
3283 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3284 {
3285         return param_set_uint_minmax(val, kp,
3286                         RPC_MIN_RESVPORT,
3287                         RPC_MAX_RESVPORT);
3288 }
3289
3290 static const struct kernel_param_ops param_ops_portnr = {
3291         .set = param_set_portnr,
3292         .get = param_get_uint,
3293 };
3294
3295 #define param_check_portnr(name, p) \
3296         __param_check(name, p, unsigned int);
3297
3298 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3299 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3300
3301 static int param_set_slot_table_size(const char *val,
3302                                      const struct kernel_param *kp)
3303 {
3304         return param_set_uint_minmax(val, kp,
3305                         RPC_MIN_SLOT_TABLE,
3306                         RPC_MAX_SLOT_TABLE);
3307 }
3308
3309 static const struct kernel_param_ops param_ops_slot_table_size = {
3310         .set = param_set_slot_table_size,
3311         .get = param_get_uint,
3312 };
3313
3314 #define param_check_slot_table_size(name, p) \
3315         __param_check(name, p, unsigned int);
3316
3317 static int param_set_max_slot_table_size(const char *val,
3318                                      const struct kernel_param *kp)
3319 {
3320         return param_set_uint_minmax(val, kp,
3321                         RPC_MIN_SLOT_TABLE,
3322                         RPC_MAX_SLOT_TABLE_LIMIT);
3323 }
3324
3325 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3326         .set = param_set_max_slot_table_size,
3327         .get = param_get_uint,
3328 };
3329
3330 #define param_check_max_slot_table_size(name, p) \
3331         __param_check(name, p, unsigned int);
3332
3333 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3334                    slot_table_size, 0644);
3335 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3336                    max_slot_table_size, 0644);
3337 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3338                    slot_table_size, 0644);